This action might not be possible to undo. Are you sure you want to continue?
r
X
i
v
:
q
u
a
n
t

p
h
/
0
1
0
8
1
4
3
v
1
3
1
A
u
g
2
0
0
1
Chapter 1
Classical and Quantum
Propagation of Chaos
1.1 Overview
The concept of molecular chaos dates back to Boltzmann [3], who derived the
fundamental equation of the kinetic theory of gases under the hypothesis that
the molecules of a nonequilibrium gas are in a state of “molecular disorder.”
The concept of propagation of molecular chaos is due to Kac [8, 9], who called
it “propagation of the Boltzmann property” and used it to derive the homo
geneous Boltzmann equation in the inﬁniteparticle limit of certain Markovian
gas models (see also [5, 17]). McKean [12, 13] proved the propagation of chaos
for systems of interacting diﬀusions that yield diﬀusive Vlasov equations in the
meanﬁeld limit. Spohn [16] used a quantum analog of the propagation of chaos
to derive timedependent Hartree equations for meanﬁeld Hamiltonians, and
his work was extended in [1] to open quantum meanﬁeld systems.
This article examines the relationship between classical and quantum prop
agation of chaos. The rest of this introduction reviews some ideas of quantum
probability and dynamics. Section 1.2 discusses the classical and quantum con
cepts of propagation of chaos. In Section 1.3, classical propagation of chaos
is shown to occur when quantum systems that propagate quantum molecular
chaos are suitably prepared, allowed to evolve without interference, and then
observed. Our main result is Corollary 1.3.7, which may be paraphrased as
follows:
Let O be a complete observable of a single particle, taking its values in
a countable set J, and let O
i
denote the observable O of particle i in a sys
tem of n distinguishable particles of the same species. Suppose we allow that
quantum nparticle system to evolve freely, except that we periodically measure
O
1
, O
2
, . . . , O
n
. The resulting time series of measurements is a Markov chain in
J
n
. If the sequence of nparticle dynamics propagates quantum molecular chaos,
then these derived Markov chains propagate chaos in the classical sense.
1
2 Alex Gottlieb
Results like this may be of interest to probabilists who already know some
examples of the propagation of chaos and who may be surprised to learn of
novel examples arising in quantum dynamics. An eﬀort has been made here
to expound the propagation of quantum molecular chaos for such an audience,
while the classical propagation of chaos per se is discussed only brieﬂy in Sec
tion 1.2.1. The reader is referred to [18] and [14] for two deﬁnitive surveys of
the classical propagation of chaos.
1.1.1 Quantum kinematics
In quantum theory, the state of a physical system is inherently statistical: the
state of a system o does not determine whether or not o has a given property,
but rather, the state provides only the probability that o would be found to
have that property, if we were to check for it. The properties that o might
or might not have are represented by orthogonal projectors on some Hilbert
space. If a projector P represents a property P of o, then the complementary
property NOT P is represented by I − P. The identity and zero operators I
and 0 represent the trivial properties TRUE and FALSE respectively. If P and
Q are properties whose orthogonal projectors are P and Q, the properties (P
AND Q) and (P OR Q) are deﬁned if and only if P and Q commute, in which
case PQ = QP represents (P AND Q) and P + Q−PQ represents (P OR Q).
A countable resolution of the identity is a countable family of projectors ¦P
j
¦
such that
P
j
P
j
′ = P
j
′ P
j
= 0 ; ∀j ,= j
′
and I =
j
P
j
. This represents a partition of the space of outcomes of a mea
surement on o into a countable set of elemental properties, which are mutually
exclusive and collectively exhaustive. A state ω of the system o is a function
that assigns probabilities to the properties of o, or their projections. Thus we
suppose that ω(I) = 1, ω(0) = 0 and also that
ω(I −P) +ω(P) = ω(I) = 1.
Indeed, it is rational to suppose that any state ω must satisfy
1 = ω(I) = ω
_
j
P
j
_
=
j
ω(P
j
) (1.1)
for any countable resolution of the identity ¦P
j
¦.
Having made these introductory comments, we revert to a more technical
description of the mathematical setup. Suppose the properties of a quantum
system o are represented by orthogonal projectors in B(H), the bounded oper
ators on a Hilbert space H. The statistical states (also called simply states) of
that quantum system are identiﬁed with the normal positive linear functionals
on B(H) that assign 1 to the identity operator. A positive linear functional ω
on B(H) is normal if
a∈A
ω(P
a
) = 1
Propagation of Chaos in Classical and Quantum Kinetics 3
whenever ¦P
a
¦
a∈A
is a family of commuting projectors that sum to the identity
operator (i.e., the net of ﬁnite partial sums of the projectors converges in the
weak operator topology to the identity). Normal states are in oneone corre
spondence with density operators, positive traceclass operators of trace 1. If D
is a density operator on H then A → Tr(DA) deﬁnes a normal state on B(H);
conversely, every normal state ω on B(H) is of the form ω(A) = Tr(DA) for
some density operator D. The density operators form a closed convex subset
of the traceclass operators, which is a Banach space with T = Tr([T[). The
dual of the Banach space of trace class operators is B(H) with its operator norm.
In the Heisenberg picture of quantum dynamics — where the state is con
stant while the operators corresponding to observables change — the dynamics
are given by unitarily implemented automorphisms of the bounded operators
on a Hilbert space. That is, for each τ ≥ 0 there exists a unitary operator U(τ)
such that a property represented by P at time t = 0 is represented by
P(τ) = U(τ)
∗
PU(τ) (1.2)
at time t = τ. In the Schr¨odinger picture of dynamics, the density operator D of
the quantum state changes in time while the projectors P that represent prop
erties of the system remain ﬁxed. The Schr¨odinger formulation of the dynamics
corresponding to (1.2) is
D(τ) = U(τ)D(0)U(τ)
∗
,
because, for any P ∈ B(H),
Tr[DP(τ)] = Tr[D U(τ)
∗
PU(τ)] = Tr[U(τ)DU(τ)
∗
P] = Tr[D(τ)P].
The Heisenberg picture of dynamics suggests a generalization where the dynam
ics A → U
∗
AU are replaced by more general endomorphisms of B(H), namely,
by completely positive maps φ : B(H) −→ B(H) such that φ(I) = I. These
endomorphisms include the automorphisms A → U
∗
AU of the Heisenberg pic
ture of quantum dynamics, but also describe maps of observables A → φ(A)
eﬀected by the intervention of measurements, randomization, and coupling to
other systems. A map φ is positive if it maps nonnegative operators to non
negative operators, and it is completely positive if φ ⊗ id
d
is positive whenever
id
d
is the identity on B(C
d
) for any ﬁnite d. Requiring φ to be positive and
unital (unit preserving) is necessary to ensure that (1.1) holds, at least for any
ﬁnite resolution of the identity. The complete positivity of φ ensures the posi
tivity of the dynamics of certain extensions of the original system o, where o is
considered together with a physically independent, ﬁnitedimensional quantum
system. To pass to the Schr¨odinger picture we must impose the further technical
requirement that the map φ be normal, i.e., φ is assumed to be such that
limφ(A
α
) = φ(A)
whenever ¦A
α
¦ is a monotone increasing net of positive operators with least
upper bound A. This way the Schr¨odinger dynamics of the normal state can
4 Alex Gottlieb
be deﬁned as the “predual” of the Heisenberg dynamics; if φ is normal then the
relation
Tr(φ
∗
(D)A) = Tr(Dφ(A)) ∀A ∈ B(H)
implicitly deﬁnes a tracepreserving map φ
∗
known as the predual of φ. In the
Schr¨odinger picture, the density operator D that describes the quantum state
undergoes the transformation D → φ
∗
(D), where φ is a normal completely
positive unital endomorphism of B(H).
The description of a quantum system evolving continuously in time requires
a normal and completely positive endomorphism of B(H) for each t > 0, to
describe the change of observables (in the Heisenberg picture) from time 0 to
time t. A quantum dynamical semigroup, or QDS, is a family ¦φ
t
¦
t≥0
of normal
completely positive (and unital) endomorphisms of the bounded operators on
some Hilbert space H, which is a semigroup (i.e., φ
0
= id and φ
t
◦ φ
s
= φ
t+s
for
s, t ≥ 0) and which has weak*continuous trajectories: for any B ∈ B(H) and
any trace class operator T
Tr(T φ
t
(B))
is continuous in t. (We will not need the continuity of trajectories in this article.)
Quantum dynamical semigroups describe the continuous change of the state
of an open quantum system whose dynamics are autonomous and Markovian.
Many models of open quantum systems are QDSs (but not all, viz. [10]). We
will use the notation (φ)
t
for the whole QDS:
(φ)
t
= ¦φ
t
¦
t≥0
. (1.3)
1.2 Classical and Quantum Molecular Chaos
1.2.1 Classical molecular chaos
Molecular chaos is a type of stochastic independence of particles manifesting
itself in an inﬁniteparticle limit.
Let Ω
n
be the nfold Cartesian power of a measurable space Ω. A probability
measure p on Ω
n
is called symmetric if
p(E
1
E
2
E
n
) = p(E
π(1)
E
π(2)
E
π(n)
)
for all measurable sets E
1
, . . . , E
n
⊂ Ω and all permutations π of ¦1, 2, . . . , n¦.
For k ≤ n, the kmarginal of p, denoted p
(k)
, is the probability measure on S
k
satisfying
p
(k)
(E
1
E
2
E
k
) = p(E
1
E
k
Ω Ω)
for all measurable sets E
1
, . . . , E
k
⊂ Ω. In the context of classical probability
theory, one deﬁnes molecular chaos as follows [18]:
Propagation of Chaos in Classical and Quantum Kinetics 5
Deﬁnition 1.2.1. Let Ω be a separable metric space. Let p be a probability
measure on Ω, and for each n ∈ N, let p
n
be a symmetric probability measure
on Ω
n
.
The sequence ¦p
n
¦ is pchaotic if the kmarginals p
(k)
n
converge weakly to
p
⊗k
as n −→ ∞, for each ﬁxed k ∈ N.
A sequence, indexed by n, of nparticle dynamics propagates chaos if molec
ularly chaotic sequences of initial distributions remain molecularly chaotic for
all time under the nparticle dynamical evolutions. In the classical contexts
[9, 12, 14, 18] the dynamics are Markovian and the state spaces are usually
taken to be separable and metrizable. Accordingly, in my dissertation [4] I
deﬁned propagation of chaos in terms of Markov transition kernels, as follows:
Deﬁnition 1.2.2 (Classical Propagation of Chaos). Let Ω be a separable
metric space. For each n ∈ N, let K
n
: Ω
n
σ(Ω
n
) −→ [0, 1] be a Markov
transition kernel which is invariant under permutations in the sense that
K
n
(x, E) = K
n
(π x, π E)
for all permutations π of the n coordinates of x and the points of E ⊂ Ω
n
. Here,
σ(Ω
n
) denotes the Borel σﬁeld of Ω
n
.
The sequence ¦K
n
¦
∞
n=1
propagates chaos if the molecular chaos of a se
quence ¦p
n
¦ entails the molecular chaos of the sequence
__
Ω
n
K
n
(x, )p
n
(dx)
_
∞
n=1
. (1.4)
The preceding formulation of the propagation of molecular chaos is techni
cally straightforward but not ﬂexible enough to cover the weaker kinds of prop
agation of chaos phenomena that occur in several applications, most notably in
the landmark derivation of the Boltzmann equation due to Lanford and King
[11]. Nonetheless, it is still worthwhile to make Deﬁnition 1.2.2. Those models
that exhibit weak propagation of chaos phenomena usually have less realistic
regularizations that propagate molecular chaos in the sense of Deﬁnition 1.2.2.
Moreover, this deﬁnition has the pleasant feature that it implies that ¦K
n
◦L
n
¦
propagates molecular chaos when ¦K
n
¦ and ¦L
n
¦ do.
1.2.2 Quantum molecular chaos
The Hilbert space of pure states of a collection of n distinguishable components
is H
1
⊗ ⊗H
n
, where H
i
is the Hilbert space for the i
th
component. The Hilbert
space for n distinguishable components of the same species will be denoted H
⊗n
.
If D
n
is a density operator on H
⊗n
, then its kmarginal, or partial trace, is a
density operator on H
⊗k
that gives the statistical state of the ﬁrst k particles.
The kmarginal may be denoted Tr
(n−k)
D
n
and deﬁned as follows: Let O be
6 Alex Gottlieb
any orthonormal basis of H. If x ∈ H
⊗k
with k < n then for any w, x ∈ H
⊗k
_
Tr
(n−k)
D
n
(w), x
_
=
y1,... ,y
n−k
∈O
¸D
n
(w ⊗y
1
⊗ ⊗y
n−k
), x ⊗ y
1
⊗ ⊗y
n−k
) .
A linear functional ω on B(H
⊗n
) is symmetric if it satisﬁes
ω(A
1
⊗ ⊗A
n
) = ω(A
π(1)
⊗A
π(2)
⊗ ⊗ A
π(n)
)
for all permutations π of ¦1, 2, . . . , n¦ and all A
1
, . . . , A
n
∈ B(H). For each
permutation π of ¦1, 2, . . . , n¦, deﬁne the unitary operator U
π
on H
⊗n
whose
action on simple tensors is
U
π
(x
1
⊗x
2
⊗ ⊗x
n
) = x
π(1)
⊗x
π(2)
⊗ ⊗x
π(n)
. (1.5)
A density operator D
n
represents a symmetric functional on B(H
⊗n
) if and only
D
n
commutes with each U
π
. Two special types of symmetric density operators
are FermiDirac densities, which represent the statistical states of systems of
fermions, and BoseEinstein densities, which represent the statistical states of
systems of bosons. BoseEinstein density operators are characterized by the
condition that D
n
U
π
= D
n
for all permutations π, and FermiDirac densities
are characterized by the condition that D
n
U
π
= sign(π)D
n
for all π.
To recapitulate, ncomponent states are given by density operators on H
⊗n
and, in the Sch¨ odinger picture, the dynamics transforms an initial state A →
Tr(DA) into a state of the form A → Tr(Dφ(A)), where φ is a normal completely
positive unital endomorphism of B(H
⊗n
). This is the context of the following
two deﬁnitions:
Deﬁnition 1.2.3. Let D be a density operator on H, and for each n ∈ N, let
D
n
be a symmetric density operator on H
⊗n
.
The sequence ¦D
n
¦ is Dchaotic in the quantum sense if, for each ﬁxed
k ∈ N, the density operators Tr
(n−k)
D
n
converge in trace norm to D
⊗k
as
n −→ ∞.
The sequence ¦D
n
¦ is quantum molecularly chaotic if it is Dchaotic in
the quantum sense for some density operator D on H.
The deﬁnitions of classical and quantum molecular chaos are somewhat
incongruous. This deﬁnition of quantum molecular chaos requires that the
marginals converge in the trace norm, whereas the notion of classical molecular
chaos used in Probability Theory requires weak convergence of the marginals. In
fact, Deﬁnition 1.2.1 of classical molecular chaos and the “commutative” version
of Deﬁnition 1.2.3 (obtained by extending that deﬁnition of quantum molecular
chaos to commutative von Neumann algebras) are not equivalent! Nonethe
less, I have chosen Deﬁnition 1.2.3 because the attractive theory of quantum
meanﬁeld kinetics presented in Section 1.2.3 favors a formulation of quantum
molecular chaos in terms of the trace norm.
Propagation of Chaos in Classical and Quantum Kinetics 7
Deﬁnition 1.2.4 (Propagation of Quantum Molecular Chaos). For each
n ∈ N, let φ
n
be a normal completely positive map from H
⊗n
to itself that ﬁxes
the identity and which commutes with permutations, i.e., such that
φ
n
(U
∗
π
AU
π
) = U
∗
π
φ
n
(A)U
π
(1.6)
for all A ∈ B(H
⊗n
) and all permutations π of ¦1, 2, . . . , n¦, where U
π
is as
deﬁned in (1.5).
The sequence ¦φ
n
¦ propagates quantum molecular chaos if the quan
tum molecular chaos of a sequence of density operators ¦D
n
¦ entails the quan
tum molecular chaos of the sequence ¦φ
n∗
(D
n
)¦.
We shall soon ﬁnd that there are interesting examples of quantum dynamical
semigroups (φ
n
)
t
with the collective property that ¦φ
n,t
¦ propagates quantum
molecular chaos for each ﬁxed t > 0. When this happens, it is convenient to say
that the sequence ¦(φ
n
)
t
¦ of QDSs propagates chaos.
Deﬁnition 1.2.5. For each n let (φ
n
)
t
be a QDS on B(H
⊗n
) that satisﬁes the
permutation condition (1.6). The sequence ¦(φ
n
)
t
¦ propagates molecular
chaos if ¦φ
n,t
¦ propagates quantum molecular chaos for every ﬁxed t > 0.
1.2.3 Spohn’s quantum meanﬁeld dynamics
There are several successful mathematical treatments of quantum meanﬁeld
dynamics. One of them, due to H. Spohn, relies upon the concept of propaga
tion of quantum molecular chaos. Spohn’s theorem [16] constitutes a rigorous
derivation of the timedependent Hartree equation for bounded meanﬁeld po
tentials.
Let V be a bounded Hermitian operator on H ⊗ H such that V U
(12)
=
U
(12)
V (y ⊗ x), representing a symmetric twobody potential. Let V
1,2
denote
the operator on H
⊗n
deﬁned by
V
1,2
(x
1
⊗x
2
⊗ ⊗x
n
) = V (x
1
⊗x
2
) ⊗x
3
⊗ ⊗x
n
, (1.7)
and for each i, j ≤ n with i < j, deﬁne V
ij
similarly, so that it acts on the
i
th
and j
th
factors of each simple tensor. This may be accomplished by setting
V
ij
= U
∗
π
V
1,2
U
π
, where π = (2j)(1i) is a permutation that puts i in the ﬁrst
place and j in the second place, and U
π
is as deﬁned in (1.5). Deﬁne the n
particle Hamiltonians H
n
as the sum of the pair potentials V
ij
, with common
coupling constant 1/n:
H
n
=
1
n
i<j
V
ij
. (1.8)
If D
n
is a state on H
⊗n
, let D
n
(t) denote the state of an nparticle system
that was initially in state D
n
and which has undergone t units of the temporal
evolution governed by the Hamiltonian (1.8):
D
n
(t) = e
−iHnt/
D
n
e
iHnt/
. (1.9)
8 Alex Gottlieb
Theorem 1.2.6 (Spohn). Suppose D is a density operator on H and ¦D
n
¦ is
a Dchaotic sequence of symmetric density operators on H
⊗n
. Then the sequence
of density operators ¦D
n
(t)¦ deﬁned in (1.8) and (1.9) is D(t)chaotic, where
D(t) is the solution at time t of the following initialvalue problem in the Banach
space of traceclass operators:
d
dt
D(t) = −
i
Tr
(n−1)
[V, D(t) ⊗D(t)]
D(0) = D. (1.10)
In other words, if H
n
is as in (1.8) and φ
n
(A) = e
iHnt/
A e
−iHnt/
then
the sequence ¦φ
n
¦ propagates quantum molecular chaos. See [16] for a short
proof.
It must be emphasized that the preceding approach does not apply to sys
tems of Fermions. In other words, Theorem 1.2.6 yields timedependent Hartree
equations but not timedependent HartreeFock equations (which include an ex
change term that enforces the Pauli Principle). The problem is that there exists
no molecularly chaotic sequence of FermiDirac states because the antisymme
try of FermiDirac states is incompatible with the factorization of molecularly
chaotic states. To derive timedependent HartreeFock equations despite this
problem is a goal of current research [15].
Spohn’s approach can be generalized to handle Hamiltonians which involve
the usual unbounded kinetic energy operators, and to handle open quantum
meanﬁeld systems. In [1], Theorem 1.2.6 is extended to systems where the free
motion of the single particle may have an unbounded selfadjoint generator and
the twoparticle generator may have the Lindblad form. The “propagation of
quantum molecular chaos” is called the “meanﬁeld property” in that article.
1.3 Classical Manifestations of the Propagation
of Quantum Molecular Chaos
A sequence ¦φ
n
¦ that propagates quantum molecular chaos mediates a variety
of instances of the propagation of classical molecular chaos.
Given a Dchaotic sequence of density operators ¦D
n
¦ one can produce a
variety of molecularly chaotic sequences ¦q
n
¦ of probability measures. For each
singleparticle measurement /, the joint probabilities q
n
of the outcome of ap
plying /to all of the particles form a molecularly chaotic sequence. Conversely,
there are ways to convert a pchaotic sequence ¦p
n
¦ of probability measures
into a Dchaotic sequence of density operators ¦D
n
¦. Suppose we are presented
with a sequence of quantum dynamics ¦φ
n
¦ that propagates quantum molecular
chaos. We can ﬁrst encode a sequence of molecularly chaotic probabilities ¦p
n
¦
as a quantum molecularly chaotic sequence of density operators ¦D
n
¦, then
allow ¦D
n
¦ to develop into a new sequence ¦φ
n∗
(D
n
)¦ under the dynamics,
and ﬁnally read the resulting quantum states by applying some singleparticle
measurement to each of the particles. This procedure converts one molecularly
Propagation of Chaos in Classical and Quantum Kinetics 9
chaotic sequence of probability measures into another, i.e., it propagates chaos
in the classical sense.
The next three sections examine the encode/develop/readprocedure that
converts quantum molecular chaos to classical molecular chaos. Finally, in Sec
tion 1.3.4, we show how to produce Markov chains that propagate chaos by
observing quantum processes that propagate chaos.
1.3.1 Generalized measurements and the reading proce
dure
Let H be a Hilbert space and (Ω, T) a measurable space. A positive operator
valued measure, or POVM, is a function X(E) from T to the positive operators
on H which is countably additive with respect to the weak operator topology:
∞
i=1
X(E
i
) = X(E)
in the weak operator topology whenever the sets E
i
∈ T are disjoint and
E = ∪E
i
. In the special case that the positive operators X(E) are selfadjoint
projections such that X(E)X(F) = 0 when E ∩ F = ∅, the POVM is just a
resolution of identity on (Ω, T). A POVM deﬁnes an aﬃne map D → p
D,X
from
density operators on H to probability measures on Ω. For any density operator
D,
p
D,X
() = Tr [DX()] (1.11)
is a countably additive probability measure on (Ω, T).
POVMs correspond to generalized Ωvalued measurements just as the spec
tral decomposition of a selfadjoint operator corresponds to a simple measure
ment. A generalized measurement is realized by performing a simple measure
ment on a composite system consisting of the system of interest an “ancilliary
system” that has been prepared to have state E, without allowing any inter
action between the system itself and the ancilliary system or the environment.
Suppose that H
a
is the Hilbert space of an ancilliary space that has been pre
pared in state E, and P(dω), ω ∈ Ω is the spectral measure belonging to an
observable O on the composite system H ⊗ H
a
. If the system of interest is in
state D when O is measured, a random ω ∈ Ω is produced, governed by the
probability law Tr[(D ⊗ E) P(dω)]. This probability law has the form (1.11),
since
Tr[(D ⊗E) P(dω)] = Tr[D Tr
(1)
((I ⊗E)P(dω))] = Tr(DX(dω))
where X(dω) is the POVM
X(dω) = Tr
(1)
((I ⊗E
1/2
)P(dω)(I ⊗E
1/2
)).
This shows that the outcome of a generalized measurement is governed by some
POVM as in (1.11). Conversely, it can be shown that any POVM arises in this
10 Alex Gottlieb
way from some conceivable (but perhaps impracticable) generalized measure
ment of the type we have just described [7].
The following lemma has a straightforward proof, which we omit.
Lemma 1.3.1. Let Ω be a separable metric space with Borel σﬁeld σ(Ω), and
let X : σ(Ω) −→ B(H) be a POVM on Ω.
For each n, let D
n
be a symmetric density operator on B(H
⊗n
), and suppose
that ¦D
n
¦ is Dchaotic in the quantum sense.
Then the sequence of probability measures ¦p
n
¦ is pchaotic, p
n
and p being
deﬁned by
p(A) = Tr(DA)
p
n
(A
1
A
2
A
n
) = Tr (D
n
X(A
1
) ⊗ ⊗X(A
n
)) .
1.3.2 Lemmas concerning the encoding procedure
The procedure for encoding probability measures as density operators depends
on our choice of a density operator valued function D(ω) on the singleparticle
space Ω, now assumed to be a separable metric space. Let us choose a function
D(ω) from Ω to the density operators on a Hilbert space H, and assume D
is continuous for technical convenience. Then we can convert a probability
measure p
n
on Ω
n
into the density operator
_
Ω
n
D(ω
1
) ⊗D(ω
2
) ⊗ ⊗D(ω
n
) p
n
(dω
1
dω
2
dω
n
)
on H
⊗n
. If ¦p
n
¦ is a pchaotic sequence of symmetric measures on Ω
n
then the
corresponding sequence of density operators is quantum molecularly chaotic —
but only in a weak sense! Think of the density operators as a subset of the
Banach space of traceclass operators. Each continuous linear functional on this
space has the form
T −→ Tr(TB)
where B is a bounded operator, for B(H) is the Banach dual of the space of
traceclass operators on H. A sequence ¦T
n
¦ of traceclass operators on H is
weakly convergent if
lim
n→∞
Tr(T
n
B) = Tr(TB)
for all B ∈ B(H).
Lemma 1.3.2. Let Ω be a separable metric space and suppose ¦p
n
¦ is a p
chaotic sequence of symmetric measures on Ω
n
. Let D(s) be a continuous func
tion from Ω to the density operators on a Hilbert space H.
Propagation of Chaos in Classical and Quantum Kinetics 11
Deﬁne
¯
D and D
n
by
¯
D =
_
Ω
D(ω)p(dω)
D
n
=
_
Ω
n
D(ω
1
) ⊗D(ω
2
) ⊗ ⊗D(ω
n
) p
n
(dω
1
dω
2
dω
n
).
(1.12)
Then, for each k, the sequence of marginals ¦Tr
(n−k)
D
n
¦ converges weakly to
¯
D
⊗k
as n −→ ∞.
Proof. The integrals in (1.12) may be deﬁned as Bochner integrals in the Ba
nach space of traceclass operators (see [6] Theorem 3.7.4). The partial trace is
a bounded operator from the Banach space of traceclass operators on H
n
to the
Banach space of traceclass operators on H
k
, so Theorem 3.7.12 of [6] implies
that
Tr
(n−k)
_
Ω
n
D(ω
1
) ⊗ ⊗D(ω
n
) p
n
(dω
1
dω
2
dω
n
)
=
_
Ω
n
D(ω
1
) ⊗ ⊗D(ω
k
) p
n
(dω
1
dω
2
dω
n
).
(1.13)
Since Bochner integration commutes with application of bounded linear func
tionals, the right hand side of (1.13) converges weakly to
_
Ω
k
D(ω
1
) ⊗D(ω
2
) ⊗ ⊗D(ω
k
) p
⊗k
(dω
1
dω
2
dω
k
) =
¯
D
⊗k
as n −→ ∞.
The preceding lemma does not conclude that ¦D
n
¦ is quantum molecularly
chaotic in the sense of our Deﬁnition 1.2.3, which requires convergence of the
partial traces in trace norm. Some additional conditions seem necessary in order
to conclude that the encoding procedure produces quantum molecular chaos.
The easiest thing to do is suppose that the quantum systems involved are ﬁnite
dimensional. This aﬀords a quick way to construct Markov transitions on any
space Ω. The opposite approach is to let the mediating quantum dynamics
occur in any Hilbert space H but to suppose that Ω is discrete. We follow these
two approaches in the next two lemmas:
Lemma 1.3.3. If H is a ﬁnite dimensional Hilbert space C
d
, and D
n
and D
are as in the statement of Lemma 1.3.2, then the sequence of states ¦D
n
¦ is
¯
Dchaotic.
Proof. From Lemma 1.3.2 we know that ¦Tr
(n−k)
D
n
¦ converges weakly to
¯
D
⊗k
as n −→ ∞. But a sequence of traceclass operators on C
d
converges in trace
norm if it converges weakly, since the Banach space of trace class operators on
12 Alex Gottlieb
on C
d
is ﬁnitedimensional. Hence, lim
n→∞
Tr
(n−k)
(D
n
) =
¯
D
⊗k
in trace norm, as
required.
Now we consider a set J with a discrete topology and σﬁeld. We note that if
the word “separable” were removed from Deﬁnition 1.2.1 of classical molecular
chaos then the following lemma would hold even for uncountable sets J:
Lemma 1.3.4. Let J be a countable set equipped with the discrete topology and
its Borel σﬁeld (so that every subset of J is measurable), and let ¦D(j)¦
j∈J
be
a family of density operators on H indexed by J.
Suppose p is a probability measure on J and ¦p
n
¦ is a pchaotic sequence of
probability measures. Deﬁne
¯
D =
j∈J
p(j)D(j)
D
n
=
(j1,... ,jn)∈J
n
p
n
(j
1
, . . . , j
n
)D(j
1
) ⊗D(j
2
) ⊗ ⊗D(j
n
).
(1.14)
Then ¦D
n
¦ is Dchaotic.
Proof. Since the series deﬁning D
n
converges in trace norm and the partial
trace operator Tr
(n−k)
is a bounded operator with respect to the trace norm, it
follows that
Tr
(n−k)
D
n
=
J
n
p
n
(j
1
, . . . , j
n
)D(j
1
) ⊗ D(j
2
) ⊗ ⊗D(j
k
)
=
J
k
p
(k)
n
(j
1
, . . . , j
k
)D(j
1
) ⊗D(j
2
) ⊗ ⊗D(j
k
).
Now ¦p
(k)
n
¦ converges weakly to p
⊗k
as n tends to inﬁnity, for ¦p
n
¦ is pchaotic.
Since a sequence of elements of ℓ
1
converges in norm if and only if it converges
weakly [2], the sequence ¦p
(k)
n
¦ converges in the ℓ
1
norm to p
⊗k
as n tends to
inﬁnity. Hence, in trace norm,
lim
n→∞
Tr
(n−k)
D
n
=
J
k
p(j
1
)p(j
2
) p(j
k
) D(j
1
) ⊗D(j
2
) ⊗ ⊗D(j
k
)
=
¯
D
⊗k
,
proving that ¦D
n
¦ is
¯
Dchaotic.
1.3.3 Putting it together: encoding, developing, and read
ing
We now formulate two abstract propositions that follow from the above two
lemmas on the encoding procedure. Proposition 1.3.5 relies on Lemma 1.3.3
and is therefore limited by the hypothesis that the mediating quantum system is
Propagation of Chaos in Classical and Quantum Kinetics 13
ﬁnitedimensional. Proposition 1.3.6 is derived from Lemma 1.3.4. It allows the
quantum dynamics to take place in an arbitrary Hilbert space but requires the
measurable spaces involved to be discrete. Despite these technical restrictions
there remains a rich variety of classical examples of the propagation of chaos
residing within each instance of the propagation of quantum molecular chaos.
Proposition 1.3.5. Let Ω be a separable metric space with Borel σﬁeld σ(Ω),
let D(s) be a continuous function from Ω to the density operators on C
d
, and
let X : σ(Ω) −→ B(C
d
) be a POVM on Ω. For each n, let φ
n
be a normal
completely positive unital endomorphism of B((C
d
)
⊗n
) that satisﬁes (1.6).
Deﬁne the Markov transition kernel K
n
on Ω
n
by
K
n
((ω
1
, ω
2
, . . . , ω
n
), A
1
A
2
A
n
)
= Tr [(D(ω
1
) ⊗ ⊗D(ω
n
)) φ
n
(X(A
1
) ⊗ ⊗X(A
n
))] .
If ¦φ
n
¦ propagates quantum molecular chaos then ¦K
n
¦ propagates molecular
chaos in the classical sense.
The proof of this proposition is omitted because it follows directly from
Lemma 1.3.1 and Lemma 1.3.3. Likewise, the following proposition follows
from Lemma 1.3.1 and Lemma 1.3.4:
Proposition 1.3.6. Let H be a Hilbert space and, for each n, let φ
n
be a com
pletely positive endomorphism of B(H
⊗n
) that satisﬁes (1.6). Let J be a count
able set equipped with the discrete topology and σﬁeld, and let X : J −→ B(H)
be a POVM on J.
Deﬁne the Markov transition matrices K
n
on J
n
by
K
n
((j
1
, . . . , j
n
), (j
′
1
, . . . , j
′
n
))
= Tr [(D(j
1
) ⊗ ⊗D(j
n
))φ
n
(X(j
′
1
) ⊗ ⊗X(j
′
n
))]
If ¦φ
n
¦ propagates quantum molecular chaos then the sequence of Markov tran
sition kernels ¦K
n
¦ propagates classical molecular chaos.
1.3.4 Periodic measurement of complete observables
The periodic measurement of a complete observable of a quantum system pro
duces a Markov chain of measurement values.
Let O be a complete observable of a system, represented by a resolution of
the identity ¦[e
j
)¸e
j
[¦
j∈J
for some orthonormal basis ¦e
j
¦
j∈J
of H. Consider
a (possibly open) quantum system whose evolution is governed by a QDS (φ)
t
.
If the natural quantum evolution of this system is interrupted periodically by
the measurement of O, then the resulting random sequence of measurement
values is a Markov chain on J. To be speciﬁc, suppose the measurements of O
are performed at times 0, T, 2T, 3T, . . . but there is no other interference with
14 Alex Gottlieb
the evolution (φ)
t
. The ﬁrst measurement of O results in a random outcome,
namely, the pure state
P
ej
= [e
j
)¸e
j
[
into which the system has collapsed. (We use Dirac notation [e)¸e[ for projection
onto the span of e.) In eﬀect, measuring the observable O prepares a pure state
P
ej
and informs us of the index j ∈ J of that pure state. Having been prepared
in the state P
ej
, the system is allowed to evolve T time units under (φ)
t
. By
time T, the state of the system is φ
T∗
(P
ej
). When O is measured at time T, the
measurement produces another random j
′
∈ J, and the system collapses into
the corresponding pure state P
e
j
′
. The probability of the transition j → j
′
is
Tr(φ
T∗
(P
ej
)P
e
j
′
) = Tr(P
ej
φ
T
(P
e
j
′
)) ≡ K(j, j
′
),
deﬁning a Markov transition K(, ) from J to itself. At time T the system has
been prepared in some pure state P
e
j
′
, and a new experiment begins: the system
undergoes T time units of the evolution (φ)
t
, transforming its state from P
e
j
′
to φ
T∗
(P
e
j
′
), and then O is measured at time 2T, instantaneously forcing the
system into the random state indexed by j
′′
with probability K(j
′
, j
′′
). This
is repeated, producing a random record of the indices j
0
, j
1
, j
2
, . . . of the pure
states the system was in upon measurement at times 0, 1, 2, . . . of O. Ideally,
these successive measurement/experiments would be independent, both physi
cally and stochastically, and it is evident that the performance of those measure
ments in succession would produce a random sequence j
0
, j
1
, j
2
, . . . governed by
the (onestep) Markov transition kernel K(, ).
It will be convenient to denote by K[(φ)
t
, O, T] the Markov transition pro
duced in this way, so that
K[(φ)
t
, O, T](j, j
′
) = Tr(P
ej
φ
T
(P
e
j
′
)). (1.15)
Imagine an ncomponent system whose (distinguishable) components are
each quantum systems with the (same) Hilbert space H, and which is governed
by a QDS (φ
n
)
t
that satisﬁes the permutation condition (1.6). Let ¦e
j
¦ be
an orthonormal basis for H indexed by J, and let O be the observable that
returns the value j if the (component) system is in the pure state e
j
. The
observable O determines a resolution of the identity ¦[e
j
)¸e
j
[¦. The Hilbert
space for the ncomponent system is H
⊗n
and the state of the system is a
density operator on H
⊗n
. Let O
i
denote the observable that returns the value j
if the i
th
component in the pure state e
j
. We can imagine measuring O
i
of each
of the components because the components are distinguishable. Simultaneous
measurement of O
1
, . . . , O
n
on the ncomponent system results in a random
vector (j
′
1
, . . . , j
′
n
) ∈ J
n
of measurement values, and forces the system into the
pure state e
j
′
1
⊗ ⊗e
j
′
n
. Let O
n
denote the joint measurement (O
1
, . . . , O
n
).
Periodic measurement of O
n
results in a Markov chain of values in J
n
, since O
n
is a complete observable of the ncomponent system. The onestep transition
kernel for this Markov chain is
K[(φ
n
)
t
, O
n
, T](j, j
′
) = Tr(P
ej
1
⊗···⊗ejn
φ
T
(P
e
j
′
1
⊗···⊗e
j
′
n
)). (1.16)
Propagation of Chaos in Classical and Quantum Kinetics 15
by formula (1.15).
Corollary 1.3.7. Let (φ
n
)
t
and O
n
be as above for each n, and suppose that
¦(φ
n
)
t
¦ propagates quantum molecular chaos. Then, for each T ≥ 0, the se
quence of Markov transitions ¦K[(φ
n
)
t
, O
n
, T]¦
n∈N
propagates chaos in the clas
sical sense.
Proof. Consider the special case of Proposition 1.3.6 where
D(j) = X(j) = [e
j
)¸e
j
[.
In that case the sequence ¦L
n
¦ of Markov transitions
L
n
((j
1
, . . . , j
n
), (j
′
1
, . . . , j
′
n
)) = Tr
_
(Q
j1,... ,jn
)φ
n
(Q
j
′
1
,... ,j
′
n
)
¸
(1.17)
with Q
j1,... ,jn
= [e
j1
)¸e
j1
[ ⊗ ⊗ [e
jn
)¸e
jn
[ propagates molecular chaos. But
comparing (1.17) to (1.16) and noting that
P
ej
1
⊗···⊗ejn
=
¸
¸
e
j1
⊗ ⊗e
jn
_¸
e
j1
⊗ ⊗e
jn
¸
¸
= [e
j1
)¸e
j1
[ ⊗[e
j2
)¸e
j2
[ ⊗[e
jn
)¸e
jn
[ = Q
j1,... ,jn
shows that
L
n
= K[(φ
n
)
t
, O
n
, T].
1.4 Acknowledgements
I am indebted to William Arveson, Sante Gnerre, Lucien Le Cam, Marc Rieﬀel,
and Geoﬀrey Sewell for their advice and for their interest in this research. I
would also like to thank Rolando Rebolledo for his collaboration and support.
The author is supported by the Austrian START project “Nonlinear Schr¨odinger
and quantum Boltzmann equations.”
Bibliography
[1] R. Alicki and J. Messer. Nonlinear quantum dynamical semigroups for
manybody open systems. Journal of Statistical Physics 32 (2): 299  312,
1983.
[2] S. Banach. Th´eorie des Op´erations Lin´eaires. Chelsea Publishing Company,
New York, 1932. (See Cap. IX, Section 2)
[3] L. Boltzmann. Lectures on Gas Theory. Dover Publications, New York,
1995.
[4] A. D. Gottlieb. Markov transitions and the propagation of chaos (PhD The
sis). Lawrence Berkeley National Laboratory Report, LBNL42839, 1998.
[5] F. A. Gr¨ unbaum. Propagation of chaos for the Boltzmann equation. Archive
for Rational Mechanics and Analysis 42: 323345, 1971.
[6] E. Hille and R. S. Phillips. Functional Analysis and Semigroups. American
Mathematical Society, Providence, Rhode Island, 1957.
[7] A. S. Holevo. Statistical decision theory for quantum systems. Journal of
Multivariate Analysis 3: 337394, 1973.
[8] M. Kac. Foundations of kinetic theory. Proceedings of the Third Berkeley
Symposium on Mathematical Statistics and Probability, Vol III. University
of California Press, Berkeley, California, 1956.
[9] M. Kac. Probability and Related Topics in Physical Sciences. American
Mathematical Society, Providence, Rhode Island, 1976.
[10] D. Kohen, C. C. Marston, and D.J Tannor. Phase space approach to the
ories of quantum dissipation. Journal of Chemical Physics 107 (13): 5236
5253, 1997.
[11] O. E. Lanford III. The evolution of large classical systems. Lecture Notes
in Physics 35: 1111. Springer Verlag, Berlin, 1975.
[12] H. P. McKean, Jr. A class of Markov processes associated with nonlinear
parabolic equations. Proceedings of the National Academy of Science 56:
19071911, 1966.
16
Propagation of Chaos in Classical and Quantum Kinetics 17
[13] H. P. McKean, Jr. Propagation of chaos for a class of nonlinear parabolic
equations. Lecture Series in Diﬀerential Equations 7: 4157. Catholic Uni
versity, Washington, D.C., 1967.
[14] S. M´el´eard. Asymptotic behavior of some interacting particle systems;
McKeanVlasov and Boltzmann models. Lecture Notes in Mathematics,
1627. Springer, Berlin, 1995.
[15] Austrian START project on “Nonlinear Schr¨odinger and quantum Boltz
mann equations” directed by Norbert Mauser of the Institut f¨ ur Mathe
matik, Universit¨at Wien (norbert.mauser@univie.ac.at).
[16] H. Spohn. Kinetic equations from Hamiltonian dynamics. Review of Modern
Physics 53: 569  614, 1980. (See Theorem 5.7)
[17] A. Sznitman.
´
Equations de type de Boltzmann, spatialement homog`enes.
Zeitschrift f¨ ur Wahrscheinlichkeitstheorie und verwandte Gebiete 66: 559
592, 1984.
[18] A. Sznitman. Topics in propagation of chaos. Lecture Notes in Mathematics,
1464. Springer, Berlin, 1991.
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue reading from where you left off, or restart the preview.